CN114859339A - A Multi-target Tracking Method Based on Millimeter-Wave Radar - Google Patents

Abstract

The invention belongs to the field of radar signal processing, and specifically provides a multi-target tracking method based on a millimeter-wave radar, which can be used for multi-target tracking of a millimeter-wave radar in a close-range scene. The invention makes full use of the Doppler dimension information of the millimeter-wave radar, calculates the correlation probability between the track and the radar clustering target by using the D-CJPDA algorithm of Doppler, and uses the cascade matching as a framework to realize the data through the KM matching algorithm. At the same time, the present invention takes into account the missed detection of the target, and uses the ellipse correlation gate algorithm to perform secondary data correlation on the clutter point traces to reduce the probability of target tracking loss; After the radar measurement point traces of similar targets are replaced, the track is updated by using the Doppler PDAF filter, thereby improving the accuracy of multi-target tracking. In conclusion, the present invention has the advantage of high tracking accuracy.

Description

A Multi-target Tracking Method Based on Millimeter-Wave Radar

technical field

The invention belongs to the field of radar signal processing, and specifically provides a multi-target tracking method based on a millimeter-wave radar, which can be used for multi-target tracking of a millimeter-wave radar in a close-range scene.

Background technique

With the rapid development of radar technology and the increasing demand for multi-target tracking in modern military and civilian applications, the research on multi-target tracking technology is deepening. ), civil aspects (including mobile robots, intelligent transportation systems, intelligent security monitoring, etc.) are widely used. Today, multi-target tracking technology involves a number of discipline-related fields. According to the overall process of multi-target tracking, the start and end of the track, the setting of the tracking threshold, the maintenance of the tracking, the collection and processing of data, and the association of the collected data. Processing and so on constitute the overall process of multi-target tracking. As a key step in comparing and processing the actual path of the target with the collected data to compare the target from which the collected data came, data association becomes the most important link in the tracking process.

Compared with single-target tracking, the data processing process of multi-target tracking is more difficult and complicated; first, how to determine the number of tracking targets is a difficult problem, and it is necessary to correspond the echo data to the target one-to-one; for the actual radar system , each sensor has more or less certain errors, and at the same time, there may be many interferences in the working environment of the radar system, lack of certain prior knowledge on the tracking target, and the influence of system errors. The above factors will cause each tracked target to have a corresponding relationship with multiple measurements. Such a problem is the data association problem, which is magnified more seriously in the process of multi-target tracking, and has also become the core problem of multi-target tracking.

The solution to the data association problem was initially improved and perfected on the Bayesian criterion, including: nearest neighbor method, probabilistic data association method, multi-hypothesis tracking method, joint probabilistic data association method and empirical joint probability data association algorithm, etc. When the multi-target tracking neighborhood is introduced into computer vision, the solution to the data association problem has become rich, such as the use of network flow model, conditional random field model, bipartite graph model, etc. DeepSort, a classic algorithm for multi-target tracking based on deep learning, converts the data association problem into a weighted bipartite graph matching problem, and implements the KM algorithm to match the track and target; however, in the multi-target tracking algorithm based on millimeter-wave radar , the traditional radar target state information only has the target position information. Compared with the rich appearance information of the image, the error of the calculation data correlation index is larger; at the same time, the algorithm adopts the Bayesian criterion correlation algorithm, which will lead to an exponential increase in the computational complexity. It is difficult to apply to actual scene requirements.

SUMMARY OF THE INVENTION

The purpose of the present invention is to propose a multi-target tracking method based on millimeter-wave radar in view of the above-mentioned defects in the prior art; the present invention makes full use of the Doppler dimension information of the millimeter-wave radar, The algorithm calculates the association probability between the track and the radar clustering target, and takes the cascade matching as the framework to realize the data association through the KM matching algorithm; at the same time, the present invention takes into account the missed detection of the target, and uses the ellipse correlation gate algorithm to detect the clutter point trace. Perform secondary data association to reduce the probability of target tracking loss; and, for the successfully matched track, replace it with the radar measurement point track belonging to the same clustered target, and then use the Doppler PDAF filter to conduct navigation. Track update, thereby improving the accuracy of multi-target tracking.

To achieve the above object, the technical scheme adopted in the present invention is:

A multi-target tracking method based on millimeter-wave radar, characterized in that it comprises the following steps:

S1: For millimeter-wave radar echo signals, through signal processing and clustering algorithms, the radar clustering targets and clutter traces are obtained;

S2: Predict the track of the existing radar track, and calculate the interconnection probability between the track prediction value and the radar cluster target; and based on the Euclidean distance between the radar cluster target and the track prediction value, the interconnection probability is weighted twice , to obtain the data correlation matrix between the radar track and the clustered targets;

S3: Taking the cascade matching as the framework, aiming at the data correlation matrix between the radar track and the clustering target, the matched radar track and the clustering target are obtained through the KM matching algorithm;

S4: For the unmatched track, the ellipse correlation gate algorithm based on the chi-square distribution is used to perform the secondary data correlation between the track and the clutter point track, and the matching track and radar target are obtained;

S5: For all matched tracks, the Kalman filter algorithm using Doppler is used to update the track;

S6: Generate a new temporary track for the unmatched radar clustering targets, and delete tracks whose latest update time exceeds the preset threshold Th3.

Further, in step S1, the specific process is:

S1.1: Parse the radar sampling data into processable echo data: convert the binary sampling data into decimal echo data, and obtain the range-Doppler spectrum through range-Doppler FFT transformation;

S1.2: Perform non-coherent accumulation on the distance-Doppler spectrum modulo of the multi-element receiving antenna, and perform two-dimensional constant false alarm detection to obtain the radar measurement point trace and its distance and speed;

S1.3: For the range-Doppler spectrum of the multi-element receiving antenna, the azimuth angle corresponding to each radar measurement point trace is estimated through the multiple signal classification algorithm;

S1.4: For each radar measurement point trace, calculate the position of the point trace in the Cartesian coordinate system according to its distance and azimuth angle, and use the DBSCAN clustering algorithm to condense the point traces of all radar measurement point traces after conversion, and obtain the radar Cluster target and clutter traces.

Further, in step S2, the specific process is:

S2.1: Based on the existing radar track, the predicted value of the track at the current moment is calculated by the Kalman prediction equation, and the latest update time of the track is updated;

S2.2: Using the unbiased measurement conversion technology to convert the state parameters of the radar clustering target to the Cartesian coordinate system;

S2.3: Calculate the residual and covariance matrix between the radar cluster target and the track prediction value;

S2.4: Calculate the Gaussian likelihood function value between the radar cluster target and the track prediction value from the residual and covariance matrix;

S2.5: Calculate the Euclidean distance between the radar cluster target and the track prediction value;

S2.6: According to the D-CJPDA algorithm formula, calculate the interconnection probability between the track prediction value and the clustering target from the Gaussian likelihood function value, specifically: the interconnection probability β _j ^{t between the radar clustering target j and the track prediction value t} for:

m为雷达聚类目标数量，T为已有雷达航迹数量，

为雷达聚类目标j与航迹预测值t间的高斯似然函数值，B为预设常数；in,

m is the number of radar clustering targets, T is the number of existing radar tracks,

is the Gaussian likelihood function value between the radar clustering target j and the track prediction value t, and B is a preset constant;

S2.7: According to the Euclidean distance between the radar clustering target and the track prediction value, the interconnection probability is secondarily weighted to obtain the data correlation matrix between the radar track and the clustering target; the second weighting is specifically:

为二次加权后雷达聚类目标j与航迹预测值t间的互联概率，

表示权重；

r_i ^t分别表示雷达聚类目标j与航迹预测值的欧式距离、雷达聚类目标i与航迹预测值的欧式距离。in,

is the interconnection probability between the radar clustering target j and the track prediction value t after secondary weighting,

express weight;

r _i ^t respectively represent the Euclidean distance between the radar cluster target j and the track prediction value, and the Euclidean distance between the radar cluster target i and the track prediction value.

Further, in step S3, the specific process is:

S3.1 performs the following steps for each latest update time in ascending order of the latest update time:

S3.1.1: For the reliable track with the same latest update time, take out the data correlation matrix corresponding to the reliable track in the data correlation matrix between the radar track and the clustered target, and perform normalization processing;

S3.1.2: Use the KM matching algorithm to process the normalized data association matrix, and perform data association on the reliable track, that is, the matching between the track and the clustering target;

S3.1.3: For a reliable track that has been successfully matched, determine whether the Euclidean distance between the clustered target and the predicted value of the track is less than the preset threshold Th1, if so, determine that the matching is successful, and delete the matching target from the radar clustering target ;

S3.2: For the unmatched reliable track and temporary track, perform data association with the remaining radar cluster targets through the KM matching algorithm to obtain the matched track and radar cluster target.

Further, in step S4, the secondary data association is specifically: calculating the Mahalanobis distance between the clutter point track and the unmatched track, and comparing it with the corresponding threshold of the chi-square distribution with a degree of freedom of 3 and a confidence of 0.97, If the threshold is met, the data association is successful.

Further, in step S5, the specific process is:

S5.1: For all successfully matched clustering targets, replace them with the radar observation point traces belonging to the clustering targets;

S5.2: For each successfully matched radar track, update the corresponding radar observation point track by using the Doppler Probabilistic Data Association Filter (PDAF);

S5.3: Update the track quality parameter and the latest update time of the track, and determine the temporary track: if the track quality parameter exceeds the preset threshold Th2, the track will be upgraded to a reliable track.

The beneficial effects of the present invention are:

The present invention provides a multi-target tracking method based on millimeter-wave radar. The present invention firstly uses the D-CJPDA algorithm using Doppler to calculate the correlation probability between the track and the radar clustering target under the cascade matching framework, and calculates the correlation probability between the track and the radar clustering target through KM The matching algorithm and the ellipse correlation gate algorithm perform multiple data associations, which improves the accuracy of data association and reduces the impact of missed target detection on target tracking. At the same time, the PDAF filter using Doppler is used to update the track, which improves the accuracy of multi-target tracking. In this way, the prediction and tracking of multiple target trajectories can be realized in a more complex environment. In conclusion, the present invention has the advantage of high tracking accuracy.

Description of drawings

FIG. 1 is a flow chart of the multi-target tracking method of the millimeter wave radar in the present invention.

FIG. 2 is a trajectory diagram formed by multi-target tracking simulation in an embodiment of the present invention.

FIG. 3 is a trajectory diagram formed by actual measurement of multi-target tracking in an embodiment of the present invention.

Detailed ways

The present invention will be described in detail below according to the accompanying drawings and preferred embodiments to make the purpose and effect of the present invention clearer and more complete; wherein, the definitions related to abbreviations and key terms are as follows:

D-CJPDA (Distance-Cheap Joint Probability Data Association, distance-weighted empirical joint probability data association),

PDA (Probability Data Association, joint probability data association),

KM (Kuhn-Munkres, Weighted Hungarian Matching Algorithm).

This embodiment provides a multi-target tracking method based on a millimeter-wave radar, the process of which is shown in FIG. 1 , and specifically includes the following steps:

S1: For millimeter-wave radar echo signals, through signal processing and clustering algorithms, the radar clustering targets and clutter traces are obtained; the details are as follows:

S1.1: Parse the radar sampling data into processable echo data: convert the binary sampling data into decimal echo data, and obtain the range-Doppler spectrum through range-Doppler FFT transformation;

S1.2: Perform non-coherent accumulation on the distance-Doppler spectrum modulo of the multi-element receiving antenna, and perform two-dimensional constant false alarm detection to obtain the radar measurement point trace and its distance and speed;

S1.3: For the range-Doppler spectrum of the multi-element receiving antenna, the azimuth angle corresponding to each radar measurement point trace is estimated through the multiple signal classification algorithm;

S1.4: For each radar measurement point trace, calculate the position of the point trace in the Cartesian coordinate system according to its distance and azimuth angle, and use the DBSCAN clustering algorithm to condense the point traces of all radar measurement point traces after conversion, and obtain the radar Clustering targets and clutter traces;

S2: Predict the track of the existing radar track, and calculate the interconnection probability between the track prediction value and the radar clustering target; the details are as follows:

S2.1: Based on the existing radar track, the predicted value of the track at the current moment is calculated by the Kalman prediction equation, and the latest update time of the track is updated;

S2.2: Using the unbiased measurement conversion technology to convert the state parameters of the radar clustering target to the Cartesian coordinate system;

S2.3: Calculate the residual and covariance matrix between the radar cluster target and the track prediction value;

S2.4: Calculate the Gaussian likelihood function value between the radar cluster target and the track prediction value from the residual and covariance matrix;

S2.5: Calculate the Euclidean distance between the radar cluster target and the track prediction value;

S2.6: According to the D-CJPDA algorithm formula, calculate the interconnection probability between the track prediction value and the clustering target from the Gaussian likelihood function value;

为：Specifically: the interconnection probability of radar clustering target j and track prediction value t

for:

in,

为雷达聚类目标j与航迹预测值t间的高斯似然函数值；v_j(k)和S(k)为k时刻对应的残差和协方差矩阵；B为预设常数，用以防止分母为零，一般设置为极小的正常数；Among them, m is the number of radar cluster targets, T is the number of existing radar tracks,

is the Gaussian likelihood function value between the radar clustering target j and the track prediction value t; v _j (k) and S (k) are the residual and covariance matrices corresponding to time k; B is a preset constant for To prevent the denominator from being zero, it is generally set to a very small positive number;

S2.7: According to the Euclidean distance between the radar clustering target and the track prediction value, the interconnection probability is weighted twice to obtain the data correlation matrix between the radar track and the clustering target;

Specifically:

为二次加权后雷达聚类目标j与航迹预测值t间的互联概率，

表示权重；

r_i ^t分别表示雷达聚类目标j与航迹预测值的欧式距离、雷达聚类目标i与航迹预测值的欧式距离；in,

is the interconnection probability between the radar clustering target j and the track prediction value t after secondary weighting,

express weight;

r _i ^t respectively represent the Euclidean distance between the radar cluster target j and the track prediction value, and the Euclidean distance between the radar cluster target i and the track prediction value;

S3: Taking the cascade matching as the framework, aiming at the data correlation matrix between the radar track and the clustered target, the matched radar track and the clustered target are obtained through the KM matching algorithm; the details are as follows:

S3.1 performs the following steps for each latest update time in ascending order of the latest update time:

S3.1.1: For the reliable track with the same latest update time, take out the data correlation matrix corresponding to the reliable track in the data correlation matrix between the radar track and the clustered target, and perform normalization processing;

S3.1.2: Use the KM matching algorithm to process the normalized data association matrix, and perform data association on the reliable track, that is, the matching between the track and the clustering target;

S3.1.3: For a reliable track that has been successfully matched, determine whether the Euclidean distance between the clustering target and the predicted value of the track is less than the preset threshold Th1 (in this embodiment, set to 0.6m), if yes, then determine that the matching is successful, and remove the matching target from the radar clustering target;

S3.2: For the unmatched reliable track and temporary track, perform data association with the remaining radar cluster targets through the KM matching algorithm to obtain the matched track and radar cluster target;

S4: For the unmatched track, the ellipse correlation gate algorithm based on the chi-square distribution is used to perform the secondary data association between the track and the clutter point track, and the matched track and radar target are obtained; the secondary data association is specifically: calculation The Mahalanobis distance between the clutter point trace and the unmatched track is compared with the corresponding threshold of the chi-square distribution with 3 degrees of freedom and 0.97 confidence. If the threshold is met, the data association is successful;

S5: For all matched tracks, the Kalman filter algorithm using Doppler is used to update the track, as follows:

S5.1: For all successfully matched clustering targets, replace them with the radar observation point traces belonging to the clustering targets;

S5.2: For each successfully matched radar track, update the corresponding radar observation point track by using the Doppler Probabilistic Data Association Filter (PDAF);

S5.3: Update the track quality parameter and the latest update time of the track, and determine the temporary track: if the track quality parameter exceeds the preset threshold Th2 (in this embodiment, set to 3), then the track Upgrade to reliable track;

S6: Generate a new temporary track for the unmatched radar clustering targets, and delete tracks whose latest update time exceeds the preset threshold Th3 (in this embodiment, set to 30).

Below in conjunction with the test, further illustrate the beneficial effects of the present invention:

1. Test conditions:

In this embodiment, the adopted millimeter-wave radar is NXP MR3003 millimeter-wave radar, and the millimeter-wave radar host computer is used for data sampling; the used transmit antenna gain is 15dBm, and the receive antenna gain is 14dBm; the carrier frequency used by the millimeter wave radar It is 76500MHz, the longest detection distance is 20m, and the maximum detection speed is 25km/h.

2. Test content:

In this embodiment, a simulation experiment is first performed. It is assumed that there are three targets with relatively close initial distances, and the uniform linear motion is carried out along the radar azimuth angle of -0.12, 0, and 0.12 radians. It is assumed that three two-by-two cross motions are set on the tracking plane. target, respectively set the initial positions as X ₁ (0)=[-4.0m, 0.3m/s, 0.0m, 1.3m/s] ^T , X ₂ (0)=[-0.0m, 0.0m/s ,0.0m,1.3m/s] ^T , X ₃ (0)=[4.0m, 0.5m/s, 0.0m, 1.3m/s] ^T , 160 consecutive frames for a total of 16s to perform multi-target tracking, and obtain multi-target tracking The simulation diagram is shown in Figure 2. It can be seen from the simulation results that the present invention can accurately identify the tracks of the three targets, and no track jump occurs during the track tracking process, and it appears that the present invention can achieve stable tracking in a relatively complex environment.

In this embodiment, the self-made data set is used for actual measurement and verification, and the radar data is sampled by the host computer platform. The data set evaluation indicators are MOTA, FN, FP, and IDs four evaluation indicators; ratio to compare indicators. The actual measurement scene is three pedestrians moving back and forth, and a total of 163 frames takes 16.3s. The multi-target tracking trajectory diagram is shown in Figure 3, and the evaluation indicators are shown in Table 1. It can be seen from the table that the millimeter wave radar of the present invention tracks the ID switching times. Remarkably reduced, it seems that the present invention adopts the improved data association algorithm to greatly improve the accuracy of target tracking, and at the same time, through the secondary clustering of clutter traces, the possibility of missed detection of targets caused by the clustering algorithm is reduced, thereby greatly improving the accuracy of target tracking. Goal tracking metrics.

Table 1: the evaluation index comparison table of the present invention and the comparative example

The above descriptions are only specific embodiments of the present invention, and any feature disclosed in this specification, unless otherwise stated, can be replaced by other equivalent or alternative features with similar purposes; all the disclosed features, or All steps in a method or process, except mutually exclusive features and/or steps, may be combined in any way.

Claims (6)

1. A multi-target tracking method based on a millimeter wave radar is characterized by comprising the following steps:

s1: obtaining radar clustering targets and clutter point traces for millimeter wave radar echo signals through signal processing and clustering algorithms;

s2: performing track prediction on the existing radar track, and calculating the interconnection probability between the track prediction value and a radar clustering target; carrying out secondary weighting on the interconnection probability based on the Euclidean distance between the radar clustering target and the track predicted value to obtain a data association matrix between the radar track and the clustering target;

s3: obtaining matched radar tracks and clustering targets by a KM (K-K matching) algorithm according to a data association matrix between the radar tracks and the clustering targets by taking cascade matching as a framework;

s4: for unmatched flight tracks, performing secondary data association of the flight tracks and clutter point tracks by adopting an elliptic correlation gate algorithm based on chi-square distribution to obtain matched flight tracks and radar targets;

s5: performing track updating on all matched tracks by adopting a Doppler Kalman filtering algorithm;

s6: and generating a new temporary track for the unmatched radar clustering targets, and deleting the track with the latest updating time exceeding a preset threshold Th 3.

2. The multi-target tracking method based on millimeter wave radar according to claim 1, wherein in step S1, the specific process is as follows:

s1.1: analyzing radar sampling data into processable echo data: converting the binary sampling data into decimal echo data, and obtaining a range-Doppler frequency spectrum through range-Doppler FFT (fast Fourier transform);

s1.2: performing non-coherent accumulation on distance-Doppler frequency spectrum modulus of the multi-element receiving antenna, and performing two-dimensional constant false alarm detection to obtain a radar measuring point trace, and the distance and the speed of the radar measuring point trace;

s1.3: estimating the azimuth angle corresponding to each radar measuring point trace for the range-Doppler frequency spectrum of the multi-element receiving antenna through a multi-signal classification algorithm;

s1.4: and aiming at each radar measurement point trace, calculating the position of the point trace in a rectangular coordinate system according to the distance and the azimuth angle of the point trace, and performing point trace aggregation on all converted radar measurement point traces by adopting a DBSCAN clustering algorithm to obtain a radar clustering target and a clutter point trace.

3. The multi-target tracking method based on millimeter wave radar according to claim 1, wherein in step S2, the specific process is as follows:

s2.1: based on the existing radar track, calculating through a Kalman prediction equation to obtain a track prediction value at the current moment, and updating the latest track updating time;

s2.2: converting the state parameters of the radar clustering targets into a rectangular coordinate system by adopting an unbiased measurement conversion technology;

s2.3: calculating residual errors and covariance matrixes between the radar clustering targets and the track predicted values;

s2.4: calculating a Gaussian likelihood function value between the radar clustering target and the track predicted value according to the residual error and the covariance matrix;

s2.5: calculating the Euclidean distance between the radar clustering target and the track predicted value;

s2.6: according to a D-CJPDA algorithm formula, the interconnection probability of the track predicted value and the clustering target is calculated through the Gaussian likelihood function value, and the method specifically comprises the following steps: interconnection probability of radar clustering target j and track predicted value t

Comprises the following steps:

wherein ,

m is the number of radar clustering targets, T is the number of existing radar tracks,

a Gaussian likelihood function value between the radar clustering target j and the track predicted value t is set, and B is a preset constant;

s2.7: carrying out secondary weighting on the interconnection probability according to the Euclidean distance between the radar clustering target and the track predicted value to obtain a data association matrix between the radar track and the clustering target; the secondary weighting specifically comprises:

wherein ,

the interconnection probability between the radar clustering target j and the track predicted value t after the secondary weighting,

representing a weight;

and respectively representing the Euclidean distance between the radar clustering target j and the track predicted value and the Euclidean distance between the radar clustering target i and the track predicted value.

4. The multi-target tracking method based on millimeter wave radar according to claim 1, wherein in step S3, the specific process is as follows:

s3.1, according to the sequence of the latest updating time from small to large, aiming at each latest updating time, executing the following steps:

s3.1.1: taking out a data association matrix corresponding to the reliable track in the data association matrix between the radar track and the clustering target and carrying out normalization processing on the data association matrix aiming at the reliable track with the same latest updating time;

s3.1.2: processing the normalized data association matrix by adopting a KM matching algorithm, and performing data association on the reliable track, namely matching the track with a clustering target;

s3.1.3: judging whether the Euclidean distance between the clustering target and the track predicted value is smaller than a preset threshold Th1 or not for the successfully matched reliable track, if so, judging that the matching is successful, and deleting the matching target from the radar clustering target;

s3.2: and aiming at unmatched reliable tracks and temporary tracks, performing data association with the remaining radar clustering targets through a KM matching algorithm to obtain matched tracks and radar clustering targets.

5. The multi-target tracking method based on millimeter wave radar according to claim 1, wherein in step S4, the secondary data association specifically comprises: and calculating the Mahalanobis distance between the clutter point trace and the unmatched flight trace, comparing the Mahalanobis distance with a threshold corresponding to chi-square distribution with the degree of freedom of 3 and the confidence coefficient of 0.97, and if the Mahalanobis distance meets the threshold, successfully associating the data.

6. The multi-target tracking method based on millimeter wave radar according to claim 1, wherein in step S5, the specific process is as follows:

s5.1: replacing all successfully matched clustering targets with radar observation point traces belonging to the clustering targets;

s5.2: for each radar track successfully matched, performing track updating on a corresponding radar observation point track by using a Doppler Probability Data Association Filter (PDAF);

s5.3: updating the track quality parameter and the latest track updating time, and judging the temporary track: and if the track quality parameter exceeds a preset threshold Th2, upgrading the track into a reliable track.

Images